Verified85% confidenceFactExact time
Anthropic称Opus 5是迄今最对齐的模型，遵守宪法程度超过Sonnet 5和Fable，欺骗行为率最低
1
Sources
85%
Confidence
Medium-term (~90 days)
Relevance
8/23/2026
First Seen
Valid until: 11/21/2026
Sources
Related Entities
Related Claims
UnverifiedAnthropic的基础模型Methos是Fable 5的基础，Anthropic曾声称Methos'太危险了，不能交给大众'55% similarUnverified对抗奉承偏差的技术路径包括宪法AI（Constitutional AI，Anthropic提出）、直接偏好优化（DPO）以及诚实性评测基准如TruthfulQA54% similarUnverified为AI构建'写作宪法'（含禁用词黑名单、文风基准、禁止强行升华结尾等）可从源头干预模型的概率选择偏好54% similarUnverifiedUP主评价Fable 5在框架能力上让人惊喜，但在角色行为逻辑、前后文一致性等细节层面仍会犯小错误53% similarUnverifiedAnthropic的模型命名体系中，Opus、Sonnet、Haiku分别对应高、中、低三个能力与成本梯度52% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/788644API
curl https://kongchang.com/api/v1/knowledge/claims/788644MCP
get_claim(id=788644)